Re: presenting data models in RIF

Chris Welty wrote:
> 
> Dave Reynolds wrote:

>> The chairs were crystal clear at the last f2f that RIF is not supposed 
>> to create yet another W3C data modelling language, it even got written 
>> on the flip chart.
> 
> <chair>
> Yes, precisely so.
> 
> But the other semantic web "modeling languages" (OWL and RDFS) were not 
> designed for interchange (between other languages). RIF's primary design 
> goal is facilitating this interchange.  There seems to me to be an 
> argument in favor of providing some minimal way of interchanging class 
> hierarchies without providing another modeling language...
> 
> However, it is a slippery slope.  As we are already seeing.
> 
> So - I'd rather not call this out of scope just yet.
> </chair>

Hummm.

RIF's design goal is to facilitate interchange of rules. I'm not 
convinced that a new means of interchange of data models is necessary 
for that goal.

It is clearly a requirement that RIF rules should be able to work with 
data which follows each of the models we decide to support; but the 
rules themselves don't typically need access to the model, just to the 
data. Further there is no requirement that I know of for taking rules 
designed for one model and applying them to a different model - that 
would seem an untenable goal given the divergence of the models.

At F2F6 we voted, but didn't formally resolve, to support three data models:

  - XML documents, modeled via XML Schema
  - RDF data, modeled via RDFS/OWL
  - object structured data, modeled by UML/MOF

[To be fair, UML/MOF was not specified unambiguously for the last 
category, there was also mention of ODM (from ODMG, not OMG).]

Our formal requirements list in the UCR currently only references the 
first two of these.

Clearly these models are rather different. In the object and XML schema 
cases the type information is fixed, in RDF/OWL it is extensible (in RIF 
terms, # could reasonably appear in the conclusions of a rule). In XML 
Schema then complex types are not simply inherited - types can be 
modified by restriction as well as extension.

In each case there is a standard way to exchange the relevant models to 
enable RIF applications to perform whatever data validation, consistency 
checking they want to do. A RIF rule set that assumes a particular data 
model will need a mechanism to designate that model in its native 
format, but that has already been proposed.

Do rules themselves need to access the data model not just the data?

In the case of semantic web applications then yes, but that's not a 
problem since RDFS/OWL are themselves encoded in RDF so an RDF access 
mechanism is sufficient.

I haven't heard any use cases for accessing the model itself in the XML 
or object cases. Gary presented a case (car/lorry/vehicle) for there 
*being* a data model but the rules seemed to only need to consult the 
type of instances not the hierarchy or domain/range constraints 
themselves). I understood Paul to be saying that direct access to the 
schema from the rules would be unusual in the business rule setting.


As a way to make progress I would find it helpful to get more use cases 
from the commercial vendors on how data models are exchanged at the 
moment and why a new RIF data model interchange would help. Based on 
things like the URC document section on processing models I had been 
under the impression that the dominant approach, outside RDF/OWL, was to 
define the data model in XML Schema and don't really see how the 
existence of a new class hierarchy relation is of significant benefit in 
processing data defined in such a way.

Dave

[Unfortunately I won't be able to make the next telecon.]
-- 
Hewlett-Packard Limited
Registered Office: Cain Road, Bracknell, Berks RG12 1HN
Registered No: 690597 England

Received on Sunday, 19 August 2007 17:43:27 UTC